Methods and Apparatus for Demosaicing Images with Highly Correlated Color Channels
In one embodiment of the invention, an apparatus is disclosed including an image sensor, a color filter array, and an image processor. The image sensor has an active area with a matrix of camera pixels. The color filter array is in optical alignment over the matrix of the camera pixels. The color filter array assigns alternating single colors to each camera pixel. The image processor receives the camera pixels and includes a correlation detector to detect spatial correlation of color information between pairs of colors in the pixel data captured by the camera pixels. The correlation detector further controls demosaicing of the camera pixels into full color pixels with improved resolution. The apparatus may further include demosaicing logic to demosaic the camera pixels into the full color pixels with improved resolution in response to the spatial correlation of the color information between pairs of colors.
Latest Intuitive Surgical Operations, Inc. Patents:
- SURGICAL INSTRUMENT CABLE CONTROL AND ROUTING STRUCTURES
- COMPACT MECHANISM FOR SURGICAL TABLE MOTION
- Systems and methods for registering an instrument to an image using change in instrument position data
- System and method for assisting tool exchange
- Abdominal hernia simulation model for surgical training
The embodiments of the invention generally relate to demosaicing images to generate full color images.
BACKGROUNDMinimally invasive medical techniques are aimed at reducing the amount of extraneous tissue that is damaged during diagnostic or surgical procedures, thereby reducing patient recovery time, discomfort, and deleterious side effects. The average length of a hospital stay for a standard surgery may also be shortened significantly using minimally invasive surgical techniques. Thus, an increased adoption of minimally invasive techniques could save millions of hospital days, and millions of dollars annually in hospital residency costs alone. Patient recovery times, patient discomfort, surgical side effects, and time away from work may also be reduced with minimally invasive surgery.
To view a surgical site, an endoscopic camera with an illumination means may be inserted into a patient's body. An image sensor in the endoscopic camera captures color images of the surgical site. The small dimensions of the end of the endoscopic camera can limit the size of the image sensor down to a single integrated circuit chip with a limited number of camera pixels. It is difficult to capture high-resolution color images with a single chip with such limited dimensions.
The color images of the surgical site may be shown to a surgeon on a monitor or a display. A surgeon may want to see a magnified region of the color images captured by the image sensor. A digital zoom may be used to magnify the region. While a digital zoom may magnify regions of the images captured by the sensor, a loss of resolution or sharpness in the magnified region can occur.
It is desirable to provide a high-resolution image output to display on a display device from images captured by an image sensor.
BRIEF SUMMARYThe embodiments of the invention are summarized by the claims that follow below.
Similar reference numbers in the different drawings are associated with the same or similar elements but may have a different configuration.
DETAILED DESCRIPTIONThis detailed description describes exemplary implementations that are illustrative of the invention, and so is explanatory and not limiting. The invention is limited only by patented claims. In the drawings, some elements have been omitted to more clearly show the embodiments of the invention.
INTRODUCTIONThe embodiments of the invention include a demosaicing algorithm that can produce higher resolution full color images when a pair of color channels is highly correlated in an array of camera pixels. When the pair of color channels is highly correlated, the pair of color channels may be considered effectively equal within a fixed scale factor. For example, statistics of medical scenes (e.g., images of surgical sites inside the body) may have a high degree of spatial correlation between color channels of pixel data (e.g., green and blue color channels in an RGB color system; other color systems may be used). Thus with medical images, these color channels may be considered to change equally from pixel location to pixel location.
In accordance with an aspect of the invention, if a high correlation between first color channel information and second color channel information likely exists for a pixel of interest, and if the first color channel information is known and the second color channel information is missing (e.g., because only the first color channel information is captured at the pixel of interest's location), then a demosaicing process uses the known first color channel information to estimate missing second color channel information for the pixel of interest. Since the estimated second color channel information is based on known information at the pixel of interest's location, rather than on known information from pixel locations surrounding the pixel of interest, the estimate is more accurate than prior art processes. If a high correlation between the first color channel information and the second color channel information does not exist for a pixel of interest, then known methods are used to estimate the missing second color channel information from second color channel information in pixels surrounding the pixel of interest. A similar process may be used for instances when first and third (and fourth, etc.) color channel information is also highly correlated at the pixel of interest, or for a second highly correlated color channel pair (e.g., third and fourth) at the pixel of interest. Thus, full color channel information for each pixel of interest in an input image is produced, and when color channel information is highly correlated in many input image pixels, this process produces an output image with higher spatial resolution than output images produced only by estimating missing color information from surrounding pixels.
In accordance with an optional aspect of the invention, a color channel correlation map is created for pixel locations within an image. This map is used during the demosaicing processes described herein.
In accordance with a further aspect of the invention, once the missing second color channel information is estimated from the known first color channel information at a pixel of interest, the estimated second color channel information is further refined by using known second color channel information from pixel locations surrounding the pixel of interest. Likewise for third, etc. color channels for the pixel of interest as applicable. Thus, even higher spatial resolution output images can be generated than when only the known color channel information at the pixel of interest is used.
Generally in one embodiment of the invention, a method of demosaicing camera pixels in an array of camera pixels in response to spatial correlation is disclosed. The method may be used for improving the spatial resolution of medical images. One method includes capturing an image from a surgical site within a patient, the captured image including initial pixel data of a plurality of camera pixels; assigning single color information to the initial pixel data of each of the plurality of camera pixels to form mono-color pixel data; in an array of a plurality of arrays of camera pixels, determining if a pair of color channels are substantially correlated, and if the pair of color channels are substantially correlated then demosaicing each camera pixel in the array with its mono-color pixel data to generate full color pixel data for each camera pixel of the plurality to form a full color image with improved resolution; and outputting the full color image with the improved resolution.
Imaging Systems with Demosaic Logic and Algorithms
Referring now to
The endoscopic camera 101 includes a mechanical interface to detachably couple to a robotic arm 111 of a patient side manipulator so that it may be moved around within a surgical site of a patient. See for example, U.S. Pat. No. 6,451,027 by Cooper et al.; U.S. Pat. No. 6,779,065 by Niemeyer; U.S. Pat. No. 6,873,883 by Moll et al.; and U.S. Pat. No. 6,331,181 by Tierney et al., all of which are incorporated herein by reference. The robotic arm 111 supports the endoscopic camera 101 within the body cavity of a patient over a surgical site 180. The endoscopic camera 101 is used to capture digital images within the body cavity of the patient. The surgical site 180 includes tissue that can reflect white light to form a typical color medical image when illuminated by visible light. The endoscopic camera captures the image in the form of alternating color pixel data values, also referred to as camera pixels, that can be described as being in different color channels respectively. For example, the series of red camera pixels in different pixel locations within a frame of an image may be described as being the red color channel. An image of tissue is a typical medical image that may have a pair of color channels (e.g., green and blue color channels formed of their respective color camera pixels) with substantial spatial correlation between pixel data values at a number of different pixel locations.
The vision control cart 102 includes an illuminator 112, an image processor 114A, and a core processor 116. The vision control cart 102 may optionally include a display monitor (not shown). The endoscopic camera 101 may be coupled to the illuminator 112 to receive visible light (VL) and direct it out of its tip into the surgical site 180 to visibly illuminate tissue for capture with one or more images sensors 126L-126R. The endoscopic camera 101 captures one or more frames of medical images within the surgical site in response to the visible light (VL) and couples them into the image processor 114A. For stereo imaging, the endoscopic camera 101 is a stereo camera for concurrently capturing left and right images of the surgical site with left and right image sensors 126L-126R.
The illuminator 112 may generate the visible light (VL), a light generated in the visible electromagnetic radiation spectrum, in response to control signals 148 received from the core processor 116. The illuminator 112 may generate the visible light (VL) to capture frames of the visible images (VI) in response to the control signals.
The visible light (VL) may be coupled into the endoscopic camera 101 by one or more optical fibers or bundles of optical fibers. Similarly, the visible images (VI) of visible tissue captured by a sensor within the surgical site may coupled into the image processor 114A via an optical fiber cable or by a wire cable.
The camera pixel data, representing images of tissue within the surgical site 180, is captured by one or more image sensors 126L,126R and coupled into the image processor 114A by one or more cables 130L,130R. The camera pixel data may be alternatively assigned a visible color such as red, green, blue and/or other colors in the electro-magnetic (EM) spectrum or a wavelength in the non-visible EM spectrum (e.g., near-infra-red). This color assignment is responsive to a color mosaic filter aligned over the image sensor.
A left color mosaic filter 124L is mounted over and in optical alignment with the camera pixels in the left image sensor 126L. Left optics 122L (one or more lenses) are mounted over and in optical alignment with the left color mosaic filter 124L and the left image sensor 126L. The right color mosaic filter 124R and right optics 122R (one or more lenses) are mounted over and in optical alignment with the camera pixels in the right image sensor 126R.
The image processor 114A includes one or more processors P 120 to process the captured images and one or more storage devices (e.g., memory) M 123 to store one or more frames of image data. For stereo imaging, the image processor 114A may include a pair of processors P 120 to process left and right captured images and a pair of storage devices (e.g., memory) M 123 to store left and right image frames.
The one or more processors P 120 of the image processor 114A may execute software instructions to perform operations on the pixel data of each frame of digital image data in order to perform the image processing and display methods disclosed herein. The image processor 114A receives commands 125 from the core processor 116 and couples the images to the core processor 116 for display on a left display output device 140L and/or a right display output device 140R of the surgeon console 103 and/or a monitor (not shown) of the vision control cart 102. Alternatively or conjunctively, the core processor 116 may receive digital images and execute software instructions to perform operations on the pixel data of each frame of digital image data in order to perform the image processing and display methods disclosed herein.
The surgeon console 103 may be coupled to the core processor 116 by a fiber optic cable for high-speed communication of digital control and image information. The surgeon console 103 may include a stereo display device formed by the left display output device 140L and the right display output device 140R to display left and right stereo images to the surgeon. Alternatively, the left display output device 140L and the right display output device 140R may provide some other type of user readable output, such as a hard copy print out of photographic images.
Further information regarding minimally invasive surgical systems may be found for example in U.S. patent application Ser. No. 11/762,165, entitled MINIMALLY INVASIVE SURGICAL SYSTEM, filed by David Q. Larkin et al. on Jun. 13, 2007, published as US Patent App. Publication 2008/0065105; and U.S. Pat. No. 6,331,181, entitled SURGICAL ROBOTIC TOOLS, DATA ARCHITECTURE, AND USE, issued to Tierney et al. on Dec. 18, 2001, both of which are incorporated herein by reference.
In
Referring now to
Cables 130L,130R are coupled to the image sensors 126L, 126R to respectively send the left camera pixel data and the right camera pixel data through the shaft 113 of the endoscope out to the image processor 114A.
In
The demosaic logic 119A receives the left camera pixels over the cable 130L and generates left full color pixels 140L. The demosaic logic 119A receives the right camera pixels over the cable 130R and generates right full color pixels 140R. The core processor 116 receives the left full color pixels 140L and may assemble a plurality of them together into a frame of left full color pixels 140L to form a left image frame for display. Similarly, the core processor 116 receives the right full color pixels 140R and may assemble a plurality of them together into a frame of right full color pixels 140R to form a right image frame for display. A plurality of left and right image frames may be assembled together over time to display stereo video images.
The core processor 116 may couple to a left output device 104L and a right output device 104R of a stereo display device in the surgeon console 103. The left display pixels 140L are coupled into the left output device 104L. The right display pixels 140R are coupled to the right output device 104R. In this manner frames of images may be displayed in time on the output devices 140L,140R.
The core processor 116 may further couple to a control processor 146 of the surgeon console 103 to receive control signals to control the imaging system 100A.
Referring now to
The imaging system 100C further includes an image processor 114B. The image sensor 126 is coupled to the image processor 114B by a cable so that pixel data of the camera pixels 130 can be communicated from the image sensor 126 to the image processor 114B. The image processor 114B includes a processor P 120 to process the captured images and a storage device (e.g., memory) M 123 to store image frames for further processing.
The system 100C further includes a liquid crystal display (LCD) 154 coupled to the image processor 114B. The system 100C may optionally include a storage device 152 coupled to the image processor 114B. The storage device may store raw camera pixel data 130 and/or display pixel data 160. The image processor 114B includes demosaic logic 119B that receives the camera pixel data 130 and executes demosaicing algorithms to generate the full color pixels 140 for display on the liquid crystal display 154 or full color pixels 160 for storage into the storage device 152. Aspects of the demosaicing algorithms performed by the demosaic logic 119B are further described herein.
Medical ScenesReferring now to
Referring now to
Referring now to
The full color pixel 140 is generated from the single color camera pixel 130. In one or more embodiments of the invention, the full color pixel 140 is generated by demosaicing logic with demosaicing algorithms in response to a substantial correlation in a pair of color channels within an array of X by Y single color camera pixels. The array of camera pixels are found within the frame 126′ of camera pixels 130. Depending upon the image sensors being used, the frame of camera pixels 126′ and the frame of full color pixels 140′ may be M pixels by N pixels (M and N being variable depending upon the image sensor) to provide a high definition format.
Referring now to
Referring now to
Referring now to
The camera pixels 130A and 130B, for example, have known color levels 230A and 230B respectively. The camera pixel 130C has a color level 230C, for example. In comparison with the color level 230C of the camera pixel 130C, the camera pixels 130A and 130B may have substantially correlated color levels 230A, 230B because they change similarly in a spatial manner from camera pixel location to camera pixel location. Similarly the color levels 230H, 230I associated with the green (G) camera pixel 130H and the blue (B) camera pixel 130I may be substantially correlated because their levels change nearly the same from pixel location to pixel location.
Referring now to
As the correlation map is to be used to determine spatial correlation between a pair of color channels (e.g., red/blue, red/green, blue/green color channels), a low-resolution estimation may be used such as an interpolation algorithm or a blurring filter to estimate the missing color information for each camera pixel in the frame to complete the correlation map. An interpolation or averaging technique over neighboring camera pixels may be used to determine the missing color information for a current camera pixel. Neighboring pixels refers to any pixels that are in the local area of the pixel of interest, including a pixel that may be multiple pixels away from the given pixel of interest. Bordering pixels refers to those pixels that are immediately adjacent or touching the give pixel of interest.
Consider camera pixel 286 for example, assuming the known color information to be red. Green color information can be estimated using the green color information from the neighboring horizontal camera pixels for example. Xl neighboring camera pixels to the left and Xr neighboring camera pixels to the right may be used to interpolate or average the green color information for the current camera pixel 286. Blue color information may be estimated for the current camera pixel 286 using the blue color information from the neighboring horizontal camera pixels, for example. As alternating camera pixels have known color information for a given color, the total of neighboring camera pixels Xl and Xr should be sufficiently large to average by dividing by at least by 2. For example, five neighboring horizontal camera pixels may be chosen on the left and right to average and determine the missing color information by dividing by at least two. Typically the same color of camera pixels are used to average and determine the missing color information given the alternating nature of the color information in the camera pixels, such as shown in
Consider camera pixel 287 for example, assuming the known color information (KC) to be green. Red color information can be estimated (EC) using the known red color information from the neighboring vertical camera pixels for example. Yu neighboring camera pixels above and Yb neighboring camera pixels below may be used to interpolate or average the red color information for the current camera pixel 287. Blue color information may be estimated (EC) for the current camera pixel 286 using the blue color information from the neighboring vertical camera pixels, for example. As alternating camera pixels have known color information for a given color, the total of neighboring camera pixels Yu and Yb should be sufficiently large to average by dividing by at least by 2. This results in estimated red color information (EC), known green color information (KC), and estimated blue color information (EC), such as shown by the full color pixel 282 in
Consider camera pixel 288 for example assuming the known color (KC) information to be blue. Red color information can be estimated (EC) using the known red color information from the neighboring vertical camera pixels and neighboring horizontal camera pixels for example. Yu neighboring camera pixels above, Yb neighboring camera pixels below, Xl neighboring camera pixels to the left, and Xr neighboring camera pixels to the right may be used to interpolate or average the red color information for the current camera pixel 288. Blue color information may be estimated (EC) for the current camera pixel 288 using the blue color information from the neighboring vertical camera pixels and neighboring horizontal camera pixels, for example. As alternating camera pixels have known color information for a given color, the total number of neighboring camera pixels Yu, Yb, Xl, and Xr should be sufficiently large to average by dividing by at least by 2. This results in estimated red color information (EC), estimated green color information (EC), and known blue color information (KC), such as shown by the full color pixel 283 in
Although not shown in
Referring now to
In contrast,
For a given array of camera pixels, levels of correlation between the color channels may be determined from the correlation map. The levels of correlation may be represented as correlation coefficients or weights that are multiplied together with the known color information to determine levels of missing color information for a current camera pixel in the array.
With the color levels of at least a pair of color channels being substantially correlated, interpolation of missing colors in the camera pixel can be made from the known color information for each given camera pixel. For example, assume the red color level 240 in
Consider pixel 130E in
A demosaicing process generally takes the known color information for the camera pixel and interpolates the missing color information to form a full color pixel. Initially, pixel data for each of the camera pixels in the frame is captured and assigned single color information in response to a color mosaic filter.
Referring now to
With single color information assigned to each camera pixel, the homogeneity of the underlying image may be estimated. The homogeneity of the underlying image is useful to determine which direction (horizontal or vertical) does the neighboring pixel from a given pixel change the least. The direction with the least change may be used to select the neighbors from which to interpolate for the missing color information.
Referring now to
In
Estimating the homogeneity of an image differs from determining correlation of a pair of color channels in an array of pixels. Homogeneity is measured along a column and/or a row of camera pixels through a given camera pixel. The color of the single camera pixel is irrelevant. In contrast, correlation is measured over an M by N array of camera pixels. Moreover, the correlation is measured between color channels such that the color of the camera pixel is relevant.
After homogeneity of an image is estimated to determine the direction of least change in pixel data from a given camera pixel, color information for the given pixel may be interpolated in the direction of least change.
Referring now to
Reference is now made to
Reference is now made to
Referring now to
To form full color pixels, the missing color information associated with each single color camera pixel needs to be estimated. For example, consider the red camera pixels 130R with known red color information R. The missing red color information R′ associated with green camera pixels and the missing red color information R″ associated with blue camera pixels needs to be estimated to complete the array of red color information 601R within the red color plane 650R.
Similarly, green camera pixels 130G have known green color information G. The missing green color information G′ associated with blue camera pixels and the missing green color information G″ associated with red camera pixels need to be estimated to complete the array of green color information 601G in the green color plane 650G.
Similarly, blue camera pixels 130B have known blue color information B. The missing blue color information B′ associated with green camera pixels and the missing blue color information B″ associated with red camera pixels need to be estimated to complete the array of blue color information 601B in the blue color plane 650B.
Full color pixels are determined from the camera pixels in a demosaicing process. The estimates of the missing color information in the color planes are used with the known color information to determine spatial correlation between color channels. The spatial correlation may be used to generate a correlation map for each pixel location in the array of camera pixels.
The known color information of a given camera pixel may be used with the correlation information of the correlation map to demosaic a camera pixel into a full color camera pixel. For example, known red color information R 130R for a given camera pixel location within the array 601 may be joined with demosaic green color information Gd and demosaic blue color information Bd to form a full color pixel 140A.
Known green color information G 130G for another given camera pixel location within the array may be joined with demosaic red color information Rd and demosaic blue color information Bd to form the full color pixel 140B.
Known blue color information B 130B for another given camera pixel location within the array may be joined with the demosaic red color information Rd and the demosaic green color information Gd to form the full color pixel 140C.
Referring now to
If a pair of color channels is highly correlated, the pair of color channels is proportional to each other by a constant scale factor. In this case, the missing color information associated with the highly correlated channel can be determined directly from the known color pixel value and a scale factor. That is, for a pixel in which color information for one color channel is known and for two color channels is unknown, and the known color channel information is highly correlated to color information in one of the two unknown color channels, then the color channel information for the highly correlated but unknown color channel can be determined directly from the known color channel information and a scale factor.
As a result, the demosaicing algorithm of the present invention can estimate one or more uncorrelated color channels with fewer pixel locations in order to fill in the remaining missing color information. At the same time, the demosaicing algorithm can produce more accurate estimates for missing pixel information because known pixel data of more surrounding camera pixels can be used to estimate the one or more uncorrelated color channels.
Moreover, if the correlation between two color channels turns out to be false or not highly correlated, the demosaicing algorithm can revert back to the typical lower-resolution demosaicing algorithms. Consequently, either full color pixel information of similar quality as produced by known interpolation methods can be achieved, or full color pixel information of higher quality can be achieved using the demosaicing algorithm of the present invention.
Within the color mosaic array 601 of the single color camera pixels 130, if it is determined that a pair of color channels is highly correlated, then methods of improving the resolution for a given camera pixel may be taken. However, if there is no correlation between color channels in the mosaic array 601 of camera pixels 130, then standard methods of interpolation may be used to estimate the missing color information. Furthermore, the standard method of interpolation may be combined with the high correlation method of interpolation to further improve resolution.
Arrays 602B-604B illustrate a typical method of interpolation for missing color information in a given camera pixel 130G to complete a full color camera pixel 140B. The standard color mosaic array is used in the arrays 602B-604B throughout this interpolation process. In the array 603B, homogeneity around the given camera pixel 130G is determined to determine the least changing direction of neighboring pixels. In the array 604B, the top and bottom neighboring pixels and/or the left and right neighboring pixels are used to interpolate the missing color information Rd,Bd in the camera pixel 130G to form the full color pixel 140B. Given the camera pixel 130G was green with know color information G, red color information Rd and blue color information Bd is added into the full color pixel 140B. The green color information G of the camera pixel 130G is passed onto the full camera pixel as the green color sub-pixel information.
If there is high correlation between a pair of color channels in the array 601 of camera pixels, the pair of color channels may be treated as having substantially proportional pixel data values by a fixed scale factor. In the case of high correlation, the pixel data of the camera pixels for the pair of color channels tends to spatially change together as was discussed herein with reference to
Arrays 601, 602A, and 603C illustrate one method of interpolation to determine missing color information if there is high correlation in the array of pixels 601 between a pair of color channels. For example, with medical images (see
The effective array 602A of camera pixels, illustrates one or more color camera pixels 130 being interpolated by themselves with their own underlying known color information to determine the missing color information of its substantially correlated channel. In response to the substantial high correlation between a pair of color channels, the known color information for a given camera pixel 130 is used to generate the unknown color information of its substantially correlated channel. With a blue camera pixel B 130B for example, the missing green pixel data G′ is derived from the known blue pixel data of the blue camera pixel. The missing blue pixel data B′ of a green camera pixel 130G is derived from the known green pixel data G of the green camera pixel. With all the missing pixel data of the highly correlated channels determined by the known pixel data of its correlated channel, any remaining uncorrelated channels, for example the red camera pixel 130R, can use more neighboring pixels to enhance standard demosaicing algorithms.
The pixel data for the different colors in a multi-channel or full color pixel 140 may be referred to as sub-pixels. For example, the full color pixel 140C includes a red color sub-pixel Rd 630R, a green color sub-pixel Gd 630G, and a blue color sub-pixel B 130B. Various widths of bits may be used to represent each color. The sub-pixel data may be used as an index into a color map or color look up table of a color palette for each color channel. For example, 8 bits per sub-pixel allows a level between zero and two hundred fifty-five to select a color palette. In a true color format, the 8 bits per sub-pixel allows a color level between zero and two hundred fifty-five of each primary color red, green, and blue. A fourth sub-pixel (see sub-pixel 140X in
In the full color pixel 140C, the base information or known sub-pixel data (KC) of the blue color B of the camera pixel 130B is joined together with the estimated sub-pixel data (EC) or interpolated the red color sub-pixel 630R and the estimated sub-pixel data (EC) or interpolated green color sub-pixel 630G. Not only may a given camera pixel use its base or known pixel information to generate the missing color information, but neighboring camera pixels may also be used to further improve the resolution of interpolating a frame of given camera pixels into a frame of full color pixels for a full color image.
The effective color mosaic array 602A illustrates that the blue and green color channels of data for blue color sub-pixels and green color sub-pixels, for example, are highly correlated. The data of the missing green color sub-pixel information are estimated by being proportional by a factor of the data of the blue sub-pixel. For example, in array 602A, the missing data G′ for the green color sub-pixel 103B′ can be generated from the known blue color data of the blue color sub-pixel 103B. If instead the red color channel was highly correlated to the blue color subchannel, missing information R″ for a red color sub-pixel 130B″ can be generated from the known blue color data B of the blue color sub-pixel 103B.
The effective color mosaic arrays 603A and 604A illustrate using additional neighboring camera pixels to better interpolate the missing color information. As mentioned previously, the standard array 601 may be transformed into an effective color mosaic array 602A of camera pixels if there is a substantial correlation between a pair of color channels in the array of camera pixels. In the color mosaic array 603A, additional angles of neighboring camera pixels may be used to detect homogeneity. Not only may the horizontal and vertical camera pixels (horizontal—east and west camera pixels, vertical—north and south camera pixels) be utilized to detect homogeneity, but the camera pixels along diagonal lines (diagonal angles not ninety degree angles; e.g., forty five degree angles) where the neighboring diagonal or corner pixels (north-west, north-east, south-west, south-east camera pixels) lie may be utilized to detect homogeneity. The neighboring or adjacent camera pixels of the given camera pixel 630A may be used to further interpolate the missing color information along with the known color information of the given camera pixel. With a high degree of correlation between a pair of color channels, the neighboring pixels along the diagonal axes may have the least change in pixel level and be used to further interpolate the missing color information, along with the known color information in the given camera pixel, to generate the full color pixel.
As mentioned previously, homogeneity is often used to detect the least amount of change between adjacent neighboring pixels. With the neighboring diagonal pixels along the diagonal angles at the corners being used, there is a greater number of potential interpolation neighboring camera pixels to choose from.
The mosaic array 604A illustrates interpolation along not only the horizontal and vertical axis, but also along the diagonal angles to interpolate the missing color information from that of the given camera pixel. For example, in
Each of the interpolation methods may be selectively used to form a full color pixel or the results of each method may be combined together to try and improve the resolution in the full color pixel. An optional weighting mechanism 650 illustrated in
Referring now to
In
Each of the correlation detector 710, high-resolution demosaic logic 712H, low-resolution demosaic logic 712L, and weighting logic 714 process each camera pixel 130 within each frame of the captured image. The correlation detector 710 selectively operates over constant or varying array sizes (variable M by variable N) of camera pixels to determine if a pair of color channels is substantially correlated within the array. The correlation detector 710, the high-resolution demosaic logic 712H and the low-resolution demosaic logic 712L can operate in parallel. However, the data flow may be buffered by data buffers in order to synchronize the control signal flow and the pixel data signal flow at the weighting logic 714 to properly generate the full color pixel 140 output there-from.
The correlation detector 710 receives an array of camera pixels 130 and detects whether or not there is at least one pair of color channels that are substantially correlated together. The correlation detector 710 generates a control output indicting the level of correlation in the array between the color channels and couples it into the weighting logic 714. The control output from the correlation detector 710 indicates substantial correlation between a pair of color channels so that the weighting logic can choose to weight and use the pixel data output from the high-resolution logic over the pixel data output from the low-resolution demosaic logic.
The high-resolution demosaic logic 712H and the low-resolution demosaic logic 712L each generate the missing color information in the camera pixel 130 for the corresponding full color pixel 140. The high-resolution demosaic logic 712H performs the high-resolution demosaicing algorithm and interpolation methods shown in
The weighting logic 714 receives the known color information from the current camera pixel and multiplexes it out to the appropriate sub-pixel within the full color pixel 140. For example, if the camera pixel 130 is a blue camera pixel, the weighting logic 714 multiplexes the camera pixel data into the blue sub-pixel of the full color pixel 140. Furthermore, the weighting logic 714 may utilize the known color information of the camera pixel 130 in the generation of the missing color information (missing sub-pixel information) for the full color pixel 140.
In response to the level of correlation, the weighting logic 714 may implement weighting algorithms to weight high-resolution data HRD from the high-resolution demosaic logic 712H, low-resolution data LRD from the low-resolution demosaic logic 712L, and the known color pixel data KC of the camera pixel 130 in the generation of missing color information (MC1 and MC2) in the full color pixel 140. For example, weights W1, W2, W3 may be generated for first missing color information MC1 and weights W4, W5, W6 may be generated for second missing color information MC2 in the following weighting equations:
MC1=W1KC+W2HRD+W3LRD.
MC2=W4KC+W5HRD+W6LRD.
The weights in the missing color equations may be discrete numeric values that select the mixture of the three pieces of information (KC, HRD, LRD) that is available. Alternatively, in response to the level of correlation detected by the correlation detector 710, one or more variables may be used to generate the missing color pixel information from the known pixel data of the camera pixel 130 and the pixel data of neighboring camera pixels.
Referring now to
A control input of the variable weighted demosaic logic 712V receives the level of correlation from the correlation detector. In response to the level correlation, the variable weighted demosaic logic 712V generates the missing color information MC1,MC2 for the full color pixel 140 from the known color information KC of the camera pixel 130 in response to a variable weighting Wvi,Wv2. The following equations for the first and second missing color information are representative.
MC1=Wv1KC.
MC2=Wv2KC.
The known color information from the camera pixel 130 is coupled into the variable weighting detector 712V and the correlation detector 710. The known color information from the camera pixel 130 is passed onto the respective color sub-pixel in the full color pixel 140 by weighting it to be 100% of the known color information. For example if the camera pixel is a green camera pixel 130G, the variable weighted demosaic logic may generate 100% weight and merely pass the known green color information of the camera pixel 130G into the green sub-pixel of the full color pixel 140. The missing sub-pixel color information is generated from the known color information in the camera pixel 130 and selectively weighted by the variable weighted demosaic logic in response to the level of correlation detected by the correlation detector.
Demosaicing MethodsReferring now to
At process block 802, a first image (also referred to as a first captured image) is captured with at least a first image sensor. The first captured image includes a first frame of initial pixel data captured by a first plurality of camera pixels. The first captured image may be a medical image captured from a surgical site within a patient. A surgical site may include tissue of the patient such that the first captured image includes a plurality of pixel data representing the tissue of the patient. A surgical site may further include one or more of robotic surgical tools and body fluids. In which case, the first captured image includes a plurality of pixel data representing the one or more of robotic surgical tools and the body fluids. After capturing the first image, the process then goes to process block 804.
At process block 804, a color mosaic of color information is assigned to the initial pixel data to generate alternating single-color pixel data of known color information for each of the first plurality of camera pixels. At least two-color channels are formed by at least two colors (e.g., red, green, blue, and/or yellow) of sub-pixel data within at least two-color planes (e.g., red, green, blue, and/or yellow color planes). A color filter array aligned with the sensor assigns the mosaic of color information to the initial grey-scale pixel data of each of the plurality of camera pixels. The process then goes to process block 806.
At process block 806, missing color information is estimated for each of the first plurality of camera pixels in the first frame. The missing color information may be estimated by averaging known color information of respective neighboring camera pixels associated with the same respective color. The estimates for the missing color information may be used to form a correlation map for the frame of camera pixels. The process then goes to process block 807.
At process block 807, the first frame may be selectively partitioned into a plurality of arrays. The selection and partitioning of the first frame into a plurality of arrays may be in response to a correlation map. The process then goes to process block 808.
At process block 808, for each array of a plurality of arrays of camera pixels in the first frame, correlation information is generated. The correlation information may be annotated to the correlation map. The correlation information may used to determine if a pair of at least two-color channels are substantially correlated within each array in response to the known color information and the estimated color information for each camera pixel in the array.
A pair of color channels are highly correlated within the array if, a change in level of color information (e.g., intensity) for the pair of color channels is substantially similar (e.g., within 90% to 100%) from camera pixel to camera pixel in the array. The process goes to process block 810.
At process block 810, a determination is made if a pair of color channels is substantially correlated within an array. If a pair of color channels is substantially correlated within an array, then each camera pixel in the array can be demosaiced in response to its single-color pixel data and the correlation information. Camera pixels are demosaiced to generate full color pixel data for each camera pixel within the array and form a portion of a full color image of the captured image. With a high degree of correlation between at least a pair of color channels, the single-color pixel data of a given camera pixel may be used alone with the correlation information to demosaic the given camera pixel into a full color pixel. A missing color of the highly correlated color channel for a given camera pixel is interpolated by scaling the single-color pixel data of the given camera pixel in response to the correlation. The scaling factor may vary depending upon the level of correlation and location of the pixel within the image frame.
Spatial resolution of a missing data for a sub-pixel may be further improved by interpolating the known camera pixel data for the given location with the neighboring pixel data.
If other color channels are not highly correlated within the array, standard interpolation methods may be used to demosaic the camera pixels to generate full color pixel data. Other missing color of the given camera pixel, one that may not be highly correlated, may be determined by interpolating the neighboring pixel data of the given camera pixel. The neighboring camera pixels may include adjacent on-axis camera pixels along vertical and horizontal axes and/or adjacent off-axis camera pixels along left and right diagonal axes. In one embodiment of the invention, the estimated color may be used to form one or more sub-pixel data of the full color pixel. Similarly if in the array of camera pixels, no pair of color channels are substantially correlated, each camera pixel in the array may be demosaiced with on-axis adjacent camera pixels to generate full color pixel data for the given camera pixel but with lower resolution.
Process block 810 may be repeated for each array of camera pixels in the frame to complete the full color image of the captured image. The process then goes to process block 812.
At process block 812, the full color image of the first frame with improved resolution is output for display. If the captured image and full color image are for still images, the process may go to process block 899 and end. If the captured image and the full color image are to be repeated for video purposes, the process goes to process block 814.
At process block 814, the processes may be repeated for each of a plurality of captured images to form a plurality of frames of full color images. The plurality of frames of full color images may be output to provide full color video with improved resolution. Upon completion of capturing video or still images, the process may go to block 899.
At process block 899, the process ends.
Referring now to
At process block 902, an image from a surgical site within a patient is captured. The captured image includes a frame of pixel data captured by a plurality of camera pixels. The process then goes to process block 904.
At process block 904, single color information is alternatively assigned to the pixel data of each of the plurality of camera pixels to form a color mosaic in the frame of initial pixel data. A color mosaic of alternating single color information in a frame of pixel data is shown, for example, by the frame 126′ in
At process block 906, within an array of a plurality of arrays of camera pixels in a frame, each camera pixel in the array is demosaiced with itself by the high-resolution demosaicing logic 712H to generate high-resolution full color pixel data for each camera pixel. Each camera pixel in the array may be further demosaiced by high-resolution demosaicing logic 712H with each and every neighboring camera pixel to generate the high-resolution full color pixel data for each camera pixel. The process then goes to block 908.
At process block 908, each camera pixel in the array is demosaiced by the low-resolution demosaicing logic 712L with each and every on-axis neighboring camera pixel to generate low-resolution full color pixel data for each camera pixel, The process then goes to block 910.
At process block 910, a correlation detector 710 is used to detect spatial correlation in intensity information of the initial pixel data for a pair of color channels (e.g., red and green, red and blue, blue and green) over the camera pixels in the array. To ease detection, a correlation map of correlation information may be generated for at least one pair of color channels for the plurality of camera pixels. The process then goes to process block 912.
At process block 912, the high-resolution full color pixel data generated by the high-resolution demosaicing logic 712H and the high-resolution full color pixel data generated by the low-resolution demosaicing logic 712L is selectively combined to generate the full color pixel data for each camera pixel in the array in response to the spatial correlation detected by the correlation detector 710. The full color pixel data of the demosaicing processes may be selectively combined together by weighting the low-resolution full color pixel data and the high-resolution full color pixel data to generate the full color pixel data for each camera pixel. Alternatively, the full color pixel data of the demosaicing processes may be selectively combined together by switching between the low-resolution full color pixel data and the high-resolution full color pixel data to generate the full color pixel data for each camera pixel.
The method may repeat the processes as needed to complete the demosaicing of a frame of camera pixels. The process may also be repeated over a plurality of frames of camera pixels to demosaic the camera pixels into full color pixels and provide full color video. Two parallel process may be executed one for left images and another for right images to generate full color stereo video. The images may be output and displayed by one or more display devices. Otherwise, the process may go to process block 999.
At process block 999, the process ends.
Referring now to
At process block 1002, single color information is alternately assigned to a plurality of camera pixels. The assignment forms a color mosaic for a frame of M by N camera pixels. The process goes to process block 1004.
At process block 1004, an image is captured from a surgical site within a patient with the plurality of camera pixels. The image includes pixel data of the plurality of camera pixels. The process then goes to process block 1006.
At process block 1006, a correlation detector 710 detects the spatial correlation of intensity information between at least one pair of colors in the pixel data of an array of camera pixels. The process goes to process block 1008.
At process block 1008, each camera pixel in the array of camera pixels is variably demosaiced with its single color information and color information of neighboring camera pixels to generate full color pixel data for each camera pixel in response to the spatial correlation.
The method may repeat the processes as needed to complete the demosaicing of a frame of camera pixels. The process may also be repeated over a plurality of frames of camera pixels to demosaic the camera pixels into full color pixels and provide full color video.
Two parallel processes may also be executed, a first process (process blocks 1002-1008) for left images and a second process for right images, to generate full color stereo video. The second process may include alternately assigning single color information to a plurality of second camera pixels. This assignment forms a color mosaic for a second frame of M by N camera pixels. A second image is captured from the surgical site within the patient with the plurality of camera pixels. The second image includes second pixel data of the plurality of camera pixels. The correlation detector detects spatial correlation of intensity information between at least one pair of colors in the pixel data of a second array of camera pixels. Each camera pixel in the second array of camera pixels is variably demosaiced with its single color information and color information of neighboring camera pixels to generate full color pixel data for each camera pixel in response to the spatial correlation.
The images may be output and displayed by one or more output devices. For example, left pixel data may be output and displayed on a left display output device of a stereo display device and right pixel data may be output and displayed on a right display output device of the stereo display device.
Otherwise if the processing of images is completed, the process goes to process block 1099. At process block 1099, the process ends.
CONCLUSIONIf a pair of color channels are substantially correlated within a color mosaic image captured by a single-chip color camera, a high-resolution demosaicing algorithm can treat the pair of color channels to be proportional to each other by a scale factor when demosaicing the image into a full color image of full color pixels. When capturing medical images (see
One or more elements of the embodiments of the invention may be implemented in software so that one or more tasks may be automatically performed with a machine, such as a processor. When implemented in software, the elements of the embodiments of the invention are essentially the program instructions or code segments to perform the one or more tasks of the methods disclosed herein. For example, a machine readable media may have stored thereon instructions that when executed by a machine causes the machine to automatically perform operations including capturing a first image from a surgical site within a patient, the first image including first pixel data of a first plurality of first camera pixels; assigning single color information to the first pixel data of each of the first plurality of camera pixels; detecting first spatial correlation of intensity information between a pair of colors in the first pixel data of a first array of camera pixels; and variably demosaicing each camera pixel in the first array of camera pixels with neighboring camera pixels to generate first full color pixel data for each first camera pixel in response to the first spatial correlation.
The program instructions or code segments can be stored in a processor readable medium (e.g., memory 123) for execution by a processor, such as processor 120 shown in
While this specification includes many specifics, these should not be construed as limitations on the scope of the disclosure or of what may be claimed, but rather as descriptions of features specific to particular implementations of the disclosure. For example, primary colors of red, green, and blue (an additive color system) are described as being base camera pixel information and sub-pixel information of a full color pixel. However, the embodiments of the invention may be employed with pixels having different type of sub-pixel information such as cyan, magenta, and yellow (CMY), a subtractive color system; hue (color), saturation (chroma), and value (brightness) (HSV), a variable color property system; or YUV and its variants, for example. Furthermore, certain features that are described in this specification in the context of separate implementations may also be implemented in combination in a single implementation. Conversely, various features that are described in the context of a single implementation may also be implemented in multiple implementations, separately or in sub-combination. Moreover, although features may be described above as acting in certain combinations and even initially claimed as such, one or more features from a claimed combination may in some cases be excised from the combination, and the claimed combination may be directed to a sub-combination or variations of a sub-combination. The claimed invention is limited only by patented claims that follow below.
Claims
1-24. (canceled)
25. An apparatus comprising:
- a first image sensor having an active area with a first matrix of first camera pixels;
- a first color filter array over the first matrix of the first pixels, the first color filter array to assign one of a plurality of colors to each of the first camera pixels;
- an image processor coupled to the first image sensor to receive the first matrix of first camera pixels, the image processor including a correlation detector to detect spatial correlation of intensity information between a pair of colors in pixel data of the first matrix of first camera pixels and to control demosaicing of the first camera pixels into first tri-color pixels with improved resolution.
26. The apparatus of claim 25, wherein the image processor further includes
- variable weighted demosaicing logic coupled to the correlation detector and the first image sensor, the variable weighted demosaicing logic to demosaic the first camera pixels into the first tri-color pixels with improved resolution in response to the spatial correlation of intensity information.
27. The apparatus of claim 25, wherein the image processor further includes
- high-resolution demosaicing logic coupled to the first image sensor to receive the first camera pixels, the high-resolution demosaicing logic to demosaic the first camera pixels into first high-resolution tri-color pixels,
- low-resolution demosaicing logic coupled to the first image sensor to receive the first camera pixels, the low-resolution demosaicing logic to demosaic the first camera pixels into first low-resolution tri-color pixels, and
- weighting logic coupled to the correlation detector, the high-resolution demosaicing logic, and the low-resolution demosaicing logic, the weighting logic to selectively weight and combine the first low-resolution tri-color pixels and the first high-resolution tri-color pixels to generate the first tri-color pixels with improved resolution in response to the spatial correlation of intensity information.
28. The apparatus of claim 27, further comprising:
- a first output device coupled to the image processor to receive the first tri-color pixels, the first output device to display the first matrix of the first tri-color pixels as a color image with improved resolution.
29. The apparatus of claim 25, further comprising:
- a second image sensor having an active area with a second matrix of second camera pixels;
- a second color filter array over the second matrix of the second pixels, the second color filter array to assign one of a plurality of colors to each of the second camera pixels; and
- wherein the image processor is further coupled to the second image sensor to receive the second matrix of second camera pixels, and
- wherein the correlation detector to further detect spatial correlation of intensity information between a pair of colors in pixel data of the second matrix of second camera pixels and control demosaicing of the second camera pixels into second tri-color pixels with improved resolution.
30. The apparatus of claim 29, wherein
- the first and second camera pixels are left and right camera pixels of a stereo camera, and
- the first and second tri-color pixels are left and right tri-color pixels for a stereo display device.
31. The apparatus of claim 30, further comprising:
- a first output device coupled to the image processor to receive the first tri-color pixels,
- a second output device coupled to the image processor to receive the second tri-color pixels,
- the first and second output devices to display the first and second matrices of the first and second tri-color pixels as a stereo color image with improved resolution.
32. The apparatus of claim 25, further comprising:
- a storage device coupled to the image processor, the storage device to store the first tri-color pixels with improved resolution.
33. The apparatus of claim 32, further comprising:
- a display device coupled to the image processor to receive the first tri-color pixels, the display device to display the first matrix of the first tri-color pixels as a color image with improved resolution.
34. An apparatus comprising:
- a first display device displaying a first image captured within a body cavity formed by a first matrix of a first plurality of first display pixels;
- wherein at least one display pixel of the first plurality of first display pixels having improved resolution in response to correlation between a pair of color channels; and
- wherein the first image includes a tissue image portion.
35. The apparatus of claim 34, further comprising:
- a second display device of a stereo display displaying a second image captured within the body cavity formed by a second matrix of a second plurality of second display pixels;
- wherein at least one display pixel of the second plurality of second display pixels having improved resolution in response to the correlation between the pair of color channels; and
- wherein the second image includes the tissue image portion.
36. The apparatus of claim 34, wherein
- the first image further includes a surgical tool image portion.
37. The apparatus of claim 34, wherein
- the first and the second images further include a surgical tool image portion within the body cavity.
38. A non-transitory machine readable media have stored thereon instructions that when executed by a machine causes the machine to perform operations comprising:
- capturing a first image from a surgical site within a patient, the first image including first pixel data of a first plurality of first camera pixels;
- assigning single color information to the first pixel data of each of the first plurality of camera pixels;
- detecting first spatial correlation of intensity information between a pair of colors in the first pixel data of a first array of camera pixels; and
- variably demosaicing each camera pixel in the first array of camera pixels with neighboring camera pixels to generate first full color pixel data for each first camera pixel in response to the first spatial correlation.
Type: Application
Filed: Feb 25, 2014
Publication Date: Jun 26, 2014
Patent Grant number: 9844313
Applicant: Intuitive Surgical Operations, Inc. (Sunnyvale, CA)
Inventors: Jeffrey DiCarlo (Menlo Park, CA), David D. Scott (Oakland, CA), Wenyi Zhao (Mountain View, CA)
Application Number: 14/189,425
International Classification: A61B 1/00 (20060101); A61B 1/04 (20060101); H04N 13/02 (20060101); H04N 9/07 (20060101); H04N 9/09 (20060101);